AI search engine and MCP server for bots and agents. One Rust binary indexes code, docs, logs, JSONL/CSV and PDFs for BM25 full-text, vector and hybrid search, code search without grep, agent memory and RAG. Elasticsearch-compatible API; Corpus Hub serves hundreds of reference memory corpora.
Rust agent-memoryai-agentsbm25claude-codeclicode-searchcursordatabaseelasticsearchfull-text-searchhybrid-searchjsonllocal-firstmcpmcp-servermodel-context-protocolragrustsearch-enginevector-search
release: regenerate stale article twins; teach FC-ALERTING the rc.80 watcher
xerj-orga5a7c01856e9
diff --git a/landing/answers/does-xerj-support-s3-alerting-plugins.html b/landing/answers/does-xerj-support-s3-alerting-plugins.htmlindex 4e80afc84..71e7823db 100644--- a/landing/answers/does-xerj-support-s3-alerting-plugins.html+++ b/landing/answers/does-xerj-support-s3-alerting-plugins.html@@ -5,7 +5,7 @@<meta charset="UTF-8"><meta name="viewport" content="width=device-width, initial-scale=1.0"><title>Does XERJ support S3, alerting or custom plugins?</title>-<meta name="description" content="Since v1.0.0-rc.77 the index CAN live in an S3-compatible bucket (storage.backend = "s3": one immutable bundle object per segment family plus snapshot.json). _watcher never runs a watch; plugins are built-in.">+<meta name="description" content="Since v1.0.0-rc.77 the index CAN live in an S3-compatible bucket; since rc.78 record packs are real; since rc.80 _watcher evaluates exactly one watch shape and refuses every other with 501. Plugins stay built-in."><link rel="preconnect" href="https://fonts.googleapis.com"><link rel="preconnect" href="https://fonts.gstatic.com" crossorigin><link href="https://fonts.googleapis.com/css2?family=Big+Shoulders+Display:wght@400;700;800;900&family=Inter:wght@400;500;600;700&family=IBM+Plex+Sans:wght@400;500;600;700&family=JetBrains+Mono:wght@400;500;700&display=swap" rel="stylesheet">@@ -17,7 +17,7 @@<meta property="og:site_name" content="XERJ.ai"><meta property="og:type" content="article"><meta property="og:title" content="Does XERJ support S3, alerting or custom plugins?">-<meta property="og:description" content="Since v1.0.0-rc.77 the index CAN live in an S3-compatible bucket (storage.backend = "s3": one immutable bundle object per segment family plus snapshot.json). _watcher never runs a watch; plugins are built-in.">+<meta property="og:description" content="Since v1.0.0-rc.77 the index CAN live in an S3-compatible bucket; since rc.78 record packs are real; since rc.80 _watcher evaluates exactly one watch shape and refuses every other with 501. Plugins stay built-in."><meta property="og:url" content="https://xerj.org/answers/does-xerj-support-s3-alerting-plugins"><meta property="og:image" content="https://xerj.org/og/xerj-card.png"><meta property="og:image:width" content="1200">@@ -26,7 +26,7 @@<meta property="og:locale" content="en_US"><meta name="twitter:card" content="summary_large_image"><meta name="twitter:title" content="Does XERJ support S3, alerting or custom plugins?">-<meta name="twitter:description" content="Since v1.0.0-rc.77 the index CAN live in an S3-compatible bucket (storage.backend = "s3": one immutable bundle object per segment family plus snapshot.json). _watcher never runs a watch; plugins are built-in.">+<meta name="twitter:description" content="Since v1.0.0-rc.77 the index CAN live in an S3-compatible bucket; since rc.78 record packs are real; since rc.80 _watcher evaluates exactly one watch shape and refuses every other with 501. Plugins stay built-in."><meta name="twitter:image" content="https://xerj.org/og/xerj-card.png"><script type="application/ld+json">{@@ -60,12 +60,12 @@"@context": "https://schema.org","@type": "TechArticle","headline": "Does XERJ support S3, alerting or custom plugins?",- "description": "Since v1.0.0-rc.77 the index CAN live in an S3-compatible bucket (storage.backend = \"s3\": one immutable bundle object per segment family plus snapshot.json). _watcher never runs a watch; plugins are built-in.",+ "description": "Since v1.0.0-rc.77 the index CAN live in an S3-compatible bucket; since rc.78 record packs are real; since rc.80 _watcher evaluates exactly one watch shape and refuses every other with 501. Plugins stay built-in.","url": "https://xerj.org/answers/does-xerj-support-s3-alerting-plugins","image": "https://xerj.org/og/xerj-card.png","inLanguage": "en","datePublished": "2026-09-18",- "dateModified": "2026-09-26",+ "dateModified": "2026-09-30","author": {"@type": "Person","name": "XERJ documentation team"@@ -243,24 +243,24 @@<div class="breadcrumb"><a href="/">HOME</a><span class="sep">·</span><a href="/answers/">ANSWERS</a><span class="sep">·</span>DOES XERJ SUPPORT S3 STORAGE, ALERTING OR CUSTOM PLUGINS YET?</div><div class="kicker"><span>ANSWERS · CAPABILITIES AND LIMITS</span></div><h1>Does XERJ support S3 storage, alerting or custom plugins yet?</h1>- <div class="source-note">AUTHOR · XERJ documentation team · REVIEWED BY · XERJ engineering team · PUBLISHED · 2026-09-18 · UPDATED · 2026-09-26</div>- <p><strong>TL;DR</strong> — One of the three shipped in v1.0.0-rc.77: an index can live in an S3-compatible bucket (<code class="inline">storage.backend = "s3"</code>). Alerting and user-supplied plugins are still not implemented, and each has a surface in the code that looks closer to done than it is, so this page says exactly what exists, with the file that proves it. The authoritative list is <a href="https://github.com/xerj-org/xerj/blob/main/ROADMAP.md"><code class="inline">ROADMAP.md</code></a>.</p>+ <div class="source-note">AUTHOR · XERJ documentation team · REVIEWED BY · XERJ engineering team · PUBLISHED · 2026-09-18 · UPDATED · 2026-09-30</div>+ <p><strong>TL;DR</strong> — The scoreboard moved twice since this page was written: an index can live in an S3-compatible bucket since v1.0.0-rc.77, record corpus packs shipped in rc.78, and alerting became PARTIAL in rc.80 (<code class="inline">_watcher</code> evaluates exactly one watch shape and refuses every other loudly). User-supplied ingest plugins remain built-in-only. This page says exactly what exists, with the file that proves it. The authoritative list is <a href="https://github.com/xerj-org/xerj/blob/main/ROADMAP.md"><code class="inline">ROADMAP.md</code></a>.</p><div class="prompt-card"><span class="prompt-label">PASTE THIS TO YOUR AI AGENT — IT DOES THE REST</span>- <code>Act as a coding agent. Read https://xerj.org/llms.txt and https://github.com/xerj-org/xerj/blob/main/ROADMAP.md before you plan around XERJ. Object storage works in BOTH directions: xerj autoindex s3://bucket/prefix reads documents out of a bucket, and since v1.0.0-rc.77 storage.backend = "s3" stores the index itself in one (needs an existing storage.s3_bucket; the WAL stays local). Do not design for alerting, user-supplied ingest plugins or downloadable corpus packs: none of those is implemented. Tell the human which of their requirements falls on that list, and use only what the roadmap lists under shipping today.</code>- <button type="button" class="prompt-copy" data-copy="Act as a coding agent. Read https://xerj.org/llms.txt and https://github.com/xerj-org/xerj/blob/main/ROADMAP.md before you plan around XERJ. Object storage works in BOTH directions: xerj autoindex s3://bucket/prefix reads documents out of a bucket, and since v1.0.0-rc.77 storage.backend = "s3" stores the index itself in one (needs an existing storage.s3_bucket; the WAL stays local). Do not design for alerting, user-supplied ingest plugins or downloadable corpus packs: none of those is implemented. Tell the human which of their requirements falls on that list, and use only what the roadmap lists under shipping today.">COPY THE PROMPT</button>+ <code>Act as a coding agent. Read https://xerj.org/llms.txt and https://github.com/xerj-org/xerj/blob/main/ROADMAP.md before you plan around XERJ. Object storage works in BOTH directions: xerj autoindex s3://bucket/prefix reads documents out of a bucket, and since v1.0.0-rc.77 storage.backend = "s3" stores the index itself in one (needs an existing storage.s3_bucket; the WAL stays local). Alerting is PARTIAL (v1.0.0-rc.80: interval-triggered xerj_decide watches that index fires; every other shape 501s). Record packs are real since rc.78 (pre-indexed packs are not). Do not design for user-supplied ingest plugins or pre-indexed packs. Tell the human which of their requirements falls on that list, and use only what the roadmap lists under shipping today.</code>+ <button type="button" class="prompt-copy" data-copy="Act as a coding agent. Read https://xerj.org/llms.txt and https://github.com/xerj-org/xerj/blob/main/ROADMAP.md before you plan around XERJ. Object storage works in BOTH directions: xerj autoindex s3://bucket/prefix reads documents out of a bucket, and since v1.0.0-rc.77 storage.backend = "s3" stores the index itself in one (needs an existing storage.s3_bucket; the WAL stays local). Alerting is PARTIAL (v1.0.0-rc.80: interval-triggered xerj_decide watches that index fires; every other shape 501s). Record packs are real since rc.78 (pre-indexed packs are not). Do not design for user-supplied ingest plugins or pre-indexed packs. Tell the human which of their requirements falls on that list, and use only what the roadmap lists under shipping today.">COPY THE PROMPT</button></div><div class="prompt-card"><span class="prompt-label">RUN THIS XERJ COMMAND</span><code>curl -s -XPUT http://127.0.0.1:9200/_watcher/watch/demo -H 'content-type: application/json' -d '{"trigger":{"schedule":{"interval":"10s"}},"input":{"simple":{}},"condition":{"never":{}}}'</code><button type="button" class="prompt-copy" data-copy="curl -s -XPUT http://127.0.0.1:9200/_watcher/watch/demo -H 'content-type: application/json' -d '{"trigger":{"schedule":{"interval":"10s"}},"input":{"simple":{}},"condition":{"never":{}}}'">COPY COMMAND</button>- <span class="hero-proof">The request is accepted and the watch is stored. Nothing will ever evaluate it, and the reply says the condition was met although the watch asked for never.</span>+ <span class="hero-proof">Since rc.80 this exact body is REFUSED with a 501 naming what is not evaluated (the condition is not xerj_decide). A supported shape — interval trigger, one-index search input, xerj_decide condition, index_alert action — is accepted AND evaluated.</span></div><div class="prompt-card"><span class="prompt-label">RUN THIS XERJ COMMAND</span><code>curl -s -XGET http://127.0.0.1:9200/_watcher/watch/demo</code><button type="button" class="prompt-copy" data-copy="curl -s -XGET http://127.0.0.1:9200/_watcher/watch/demo">COPY COMMAND</button>- <span class="hero-proof">The stored body comes back unchanged. That is all this API does today.</span>+ <span class="hero-proof">The stored body comes back unchanged.</span></div><div class="prompt-card"><span class="prompt-label">RUN THIS XERJ COMMAND</span>@@ -269,8 +269,8 @@<span class="hero-proof">The percolate query is real. It matches stored queries against the document you supply, which is the piece a future detection feature builds on.</span></div><h2>Why this page exists</h2>- <p>A search engine that speaks a familiar wire protocol invites assumptions. An agent that sees a <code class="inline">_watcher</code> route may plan around it, and it should not: XERJ does not run watches. An operator who sees <code class="inline">backend = "s3"</code> in a config schema may expect a bucket to work — since v1.0.0-rc.77 it does, and this page is the one place that used to say otherwise.</p>- <p>XERJ's rule is that an input is either honoured or refused loudly. Two of the three follow that rule today and one does not (<code class="inline">_watcher</code> accepts a watch and never evaluates it — the accepted-and-ignored surface). All of them were checked against <code class="inline">main</code> by reading the named file: 2026-09-18 originally, with the S3 status re-verified on 2026-09-26 against v1.0.0-rc.77.</p>+ <p>A search engine that speaks a familiar wire protocol invites assumptions. An agent that sees a <code class="inline">_watcher</code> route may plan around full Elasticsearch watcher semantics — cron schedules, transforms, webhooks — and it should not: since v1.0.0-rc.80 XERJ evaluates exactly one watch shape and refuses the rest. An operator who sees <code class="inline">backend = "s3"</code> in a config schema may expect a bucket to work — since v1.0.0-rc.77 it does, and this page is the one place that used to say otherwise.</p>+ <p>XERJ's rule is that an input is either honoured or refused loudly. <code class="inline">_watcher</code> was the one surface that broke the rule (accepted a watch, never evaluated it); v1.0.0-rc.80 closed that — <a href="https://github.com/xerj-org/xerj/pull/1082">#1082</a>, closing <a href="https://github.com/xerj-org/xerj/issues/1062">#1062</a>. All statuses on this page were checked against <code class="inline">main</code> by reading the named file: 2026-09-18 originally, S3 re-verified 2026-09-26 against v1.0.0-rc.77, alerting and packs re-verified 2026-09-30 against v1.0.0-rc.80.</p><h2>S3 and object storage: both directions work</h2><p>This section said the opposite until 2026-09-26 — it described <code class="inline">S3Backend</code> as a local-directory simulation with no network client, and the startup refusal as permanent. That was true when it was checked (2026-09-18, before rc.77) and stopped being true in v1.0.0-rc.77. The correction is stated in the open rather than quietly rewritten.</p><p><strong>Reading documents out of a bucket works.</strong> <code class="inline">xerj autoindex s3://bucket/prefix</code> lists the prefix, streams in each object whose ETag or size changed, and indexes it with the same extractors it uses for a folder. <code class="inline">r2://</code> and any S3-compatible store behind <code class="inline">--endpoint-url</code> work the same way. The <a href="/answers/search-files-in-an-s3-bucket">S3 indexing page</a> has the commands and the request arithmetic.</p>@@ -283,40 +283,41 @@<li>the WAL stays local. <a href="https://github.com/xerj-org/xerj/blob/main/docs/OBJECT_STORAGE.md"><code class="inline">docs/OBJECT_STORAGE.md</code></a> is the full design page.</li></ul><p>One refusal remains, and it is the loud kind: with <code class="inline">storage.backend = "s3"</code> and no <code class="inline">storage.s3_bucket</code>, the server does not start. It prints that <code class="inline">storage.backend = "s3"</code> requires <code class="inline">storage.s3_bucket</code> to name an existing bucket — XERJ never creates one. That check is in <code class="inline">engine/crates/xerj-common/src/config.rs</code>.</p>- <h2>Alerting: there is none</h2>- <p>XERJ has no alerting. Two surfaces make it look otherwise.</p>- <p>The first is <code class="inline">_watcher</code>. <code class="inline">PUT /_watcher/watch/{id}</code> is accepted. The handler puts the body into an in-memory map and replies that the condition was met, whatever condition you sent. <code class="inline">GET</code> and <code class="inline">DELETE</code> work on the same map. There is no scheduler, and no code ever evaluates a stored watch. This one is an accepted-and-ignored input, and the roadmap says it must either run watches or refuse them.</p>- <p>The second is the console. It owns two system indices, <code class="inline">.xerj_alert_rules</code> and <code class="inline">.xerj_alert_fires</code>. Both have schemas and both are created at start-up. No evaluator reads the first or writes the second, so there are no alerts and no notifications from them either.</p>- <p>One piece underneath is real. The <code class="inline">percolate</code> query is a dispatched query type: you store queries as documents, and a <code class="inline">percolate</code> search returns the stored queries that match a document you supply. A future detection feature would use that as its cheap first stage. The stages after it, a typed judgment and an alert that carries a calibrated probability, are planned and do not exist.</p>+ <h2>Alerting: one real watch shape since v1.0.0-rc.80</h2>+ <p>This section said "there is none" until 2026-09-30 — and that was true when it was checked (2026-09-18): <code class="inline">PUT /_watcher/watch/{id}</code> accepted every body, replied that the condition was met, and no code ever evaluated anything. v1.0.0-rc.80 (<a href="https://github.com/xerj-org/xerj/pull/1082">#1082</a>) replaced the accepted-and-ignored surface with evaluate-or-refuse.</p>+ <p><strong>What evaluates.</strong> Exactly one watch shape: <code class="inline">trigger.schedule.interval</code> (>= 1s), <code class="inline">input.search</code> over exactly one index, <code class="inline">condition.xerj_decide</code> (a question, a positive label, <code class="inline">p_min</code>, <code class="inline">k</code>, an optional decide index) and an <code class="inline">index_alert</code> action. A per-watch background task prefilters candidates with the watch's own query (ascending <code class="inline">_seq_no</code> keyset paging, 100 per page), runs each rendered question through the real <code class="inline">/_decide</code>, and writes a fire record into <code class="inline">.xerj_alert_fires</code> when the positive label wins non-abstaining at or above <code class="inline">p_min</code>. Fire records carry the RAW decide confidence, plus <code class="inline">p_cal</code> with <code class="inline">calibrated: true</code> when the node has a fitted calibration — the threshold stays on the raw value.</p>+ <p><strong>What refuses.</strong> Everything else — cron schedules, transforms, other action types, multi-index input — is a 501 that names what is not evaluated. Malformed supported shapes are 400.</p>+ <p><strong>Honest limits.</strong> A restart does not resume watches (evaluators are spawned by <code class="inline">put_watch</code> only); the per-pass decide budget is 1000, with the cursor held below a failure so the backlog is retried, never dropped; the console's <code class="inline">.xerj_alert_rules</code> index still has no writer — watches are authored through the API, not the UI.</p>+ <p>The <code class="inline">percolate</code> query underneath is unchanged and real: you store queries as documents, and a <code class="inline">percolate</code> search returns the stored queries that match a document you supply. The watcher's prefilter is that idea, productised for one condition type.</p><h2>Custom ingest plugins: built-in only</h2><p>The ingest pipeline does run transforms on <code class="inline">_bulk</code>. They are built-in native Rust code: rename, drop, add, JSON parse, timestamp parse, PII redaction, grok and route.</p><p>The crate that holds them is named <code class="inline">xerj-wasm</code>, which suggests more than is there. Its <code class="inline">Cargo.toml</code> has no <code class="inline">wasmtime</code> dependency and no <code class="inline">wasm</code> feature. You cannot load your own module.</p><p>The refusal rule holds here. A pipeline that names a processor this build does not implement is stored as unrunnable, and every ingest through it is refused. It is never run as a shorter pipeline in silence.</p><h2>Two more that are planned, not shipped</h2><p><strong>A log-specific index mode.</strong> The <code class="inline">xerj-logs</code> crate is in the workspace and is compiled in as a dependency, but it is not wired: it has zero call sites outside its own tests. Logs you index today go through the general segment format and the ordinary aggregations. The <a href="/answers/cheap-low-volume-log-search">low-volume log search page</a> describes what does work.</p>- <p><strong>Downloadable corpus packs.</strong> A hub of signed, pre-indexed packs is on the roadmap and no code exists. The design work is the pack format: a format version that readers refuse when they do not know it, per-file checksums, a signature, and licence and provenance on every record. Three risks decide whether it ships: whether the source licence allows redistribution, whether a pack is safe to open, and whether the format can stay stable after packs are published.</p>+ <p><strong>Downloadable corpus packs — the RECORD half shipped in v1.0.0-rc.78</strong> (<a href="https://github.com/xerj-org/xerj/pull/1046">#1046</a>, <a href="https://github.com/xerj-org/xerj/pull/1048">#1048</a>; this section said "no code exists" before, which was true then). <code class="inline">xerj corpus build</code> turns a declarative recipe into a checksummed portable pack of records — a <code class="inline">format_version</code> readers refuse when unknown, licence and provenance on every record — <code class="inline">corpus add --from <pack></code> verifies the per-file SHA-256 and <code class="inline">--verify-sig</code> a detached ed25519 signature, and the <code class="inline">rust-vulns</code> pack is rebuilt, signed and published daily as a dated GitHub Release. Still unbuilt: the PRE-INDEXED half (an already-indexed bundle you mount instead of indexing records yourself) and a hub directory beyond that one pack.</p><h2>What to do instead today</h2><table><thead><tr><th>You wanted</th><th>What works now</th></tr></thead><tbody><tr><td>index data in S3</td><td>works since v1.0.0-rc.77: <code class="inline">storage.backend = "s3"</code> plus an existing <code class="inline">storage.s3_bucket</code> — segment bundles and <code class="inline">snapshot.json</code> live in the bucket, the WAL stays local</td></tr><tr><td>search documents that live in S3</td><td>works: <code class="inline">xerj autoindex s3://bucket/prefix</code> mirrors the changed objects to local disk and indexes them</td></tr>- <tr><td>an alert when a document matches</td><td>there is no alerting; run a <code class="inline">percolate</code> or an ordinary search on your own schedule and act on the result yourself</td></tr>+ <tr><td>an alert when a document matches</td><td>one shape works since rc.80: an interval watch with a <code class="inline">xerj_decide</code> condition over one index writes fires into <code class="inline">.xerj_alert_fires</code>; there is no notification delivery and no console authoring yet</td></tr><tr><td>a custom transform at ingest</td><td>one of the built-in transforms, or transform the document before you send it</td></tr>- <tr><td>a ready-made reference corpus</td><td>clone the source and run <code class="inline">xerj autoindex</code> on it</td></tr>+ <tr><td>a ready-made reference corpus</td><td><code class="inline">xerj corpus add --from</code> a record pack (the daily <code class="inline">rust-vulns</code> pack is published), or clone source and run <code class="inline">xerj autoindex</code> on it</td></tr></tbody></table><h2>What this page does not claim</h2>- <p>It does not claim a date for the two that remain — alerting and user-supplied plugins are planned, which means no code exists yet. S3 as an index home shipped in v1.0.0-rc.77; before that release the same setting refused to start, and this page said so.</p>+ <p>It does not claim more alerting than one watch shape: no cron, no transforms, no notification delivery, no watch authoring in the console, and watches do not survive a restart. Those remain planned. It does not claim pre-indexed packs or a pack hub: only record packs exist. S3 as an index home shipped in v1.0.0-rc.77; before that release the same setting refused to start, and this page said so.</p><p>It does not claim the search side is limited in the same way. Full-text search, the <code class="inline">hybrid</code> query and vector search are shipping, and the default embedder is lexical feature hashing, not a neural model. The <a href="/answers/how-xerj-combines-search">hybrid retrieval page</a> covers that.</p><p>XERJ is single-node. Everything on this page describes one process on one host.</p><h2>FAQ</h2><h3>Can XERJ store its index in S3, send alerts, or run my own ingest plugin?</h3>- <p>One of the three, since v1.0.0-rc.77: an index can live in an S3-compatible bucket (<code class="inline">storage.backend = "s3"</code>). Alerting and user-supplied plugins are still not implemented, and each has a surface in the code that looks closer to done than it is. This page says exactly what exists.</p>+ <p>S3 fully, since v1.0.0-rc.77 (<code class="inline">storage.backend = "s3"</code>); alerting PARTIALLY, since v1.0.0-rc.80 (one watch shape evaluates, everything else 501s); record corpus packs since rc.78. User-supplied ingest plugins are the one still not implemented, and the <code class="inline">xerj-wasm</code> crate name makes it look closer to done than it is. This page says exactly what exists.</p><h3>Does XERJ support S3 or object storage?</h3><p>Yes, in both directions. <code class="inline">xerj autoindex s3://bucket/prefix</code> reads documents OUT of a bucket as a source, and since v1.0.0-rc.77 <code class="inline">storage.backend = "s3"</code> stores the index itself in the bucket: one immutable ZBM1 bundle object per segment family plus a <code class="inline">snapshot.json</code> catalogue, written by a real S3-compatible client (Cloudflare R2, MinIO, AWS S3). Before rc.77 the index side refused to start on purpose — an earlier version of this page said it always would.</p><h3>Does XERJ have alerting or a working watcher?</h3>- <p>No. XERJ has no alerting. <code class="inline">_watcher</code> stores watches and no scheduler ever evaluates them, and the console alert-rule indices have schemas and no evaluator.</p>+ <p>Partially, since v1.0.0-rc.80. One watch shape is real: <code class="inline">PUT /_watcher/watch/{id}</code> with an interval trigger, a one-index search input, a <code class="inline">condition.xerj_decide</code> question and an <code class="inline">index_alert</code> action runs a background evaluator that writes fires into <code class="inline">.xerj_alert_fires</code> when the positive label wins non-abstaining at or above <code class="inline">p_min</code>. Everything else — cron, transforms, other actions, multi-index input — is refused with a 501 that names what is not evaluated. Restarts do not resume watches.</p><h3>Can I write my own ingest plugin for XERJ?</h3><p>Not yet. The ingest pipeline runs built-in native transforms only. The crate is named <code class="inline">xerj-wasm</code>, but the wasmtime backend is not in the tree.</p><h3>Why does the server refuse to start when I set the storage backend to s3?</h3>@@ -324,7 +325,7 @@<h3>Is there anything real to build a detection on today?</h3><p>Yes, one piece. The <code class="inline">percolate</code> query is a dispatched query type and matches stored queries against a document you supply. The judging and alerting stages that would sit after it do not exist.</p><h3>Can I download a pre-indexed corpus for XERJ?</h3>- <p>No. A hub of signed, pre-indexed packs is planned and no code exists. The open questions are redistribution licence, pack safety and format stability, and they decide whether it ships.</p>+ <p>Record packs, yes, since v1.0.0-rc.78: <code class="inline">xerj corpus build</code> turns a recipe into a checksummed, signable pack of records and <code class="inline">corpus add --from <pack></code> verifies it — the rust-vulns pack is published daily. PRE-INDEXED packs (an indexed bundle you mount) are still unbuilt, and there is no hub beyond that one pack.</p><h3>Where is the authoritative status?</h3><p><code class="inline">ROADMAP.md</code> in the repository. If this page and the roadmap disagree, the roadmap wins, and the disagreement is a bug worth an issue.</p><h2>Evidence</h2>@@ -332,8 +333,8 @@<li>storage.backend = "s3" stores the index in the bucket: one immutable ZBM1 bundle object per segment family plus a per-index snapshot.json catalogue; merges publish before retiring inputs; a fresh node adopts the bucket. — <code>docs/OBJECT_STORAGE.md</code></li><li>The one remaining startup refusal is the loud kind: storage.backend = "s3" requires storage.s3_bucket to name an existing bucket — XERJ never creates one. — <code>engine/crates/xerj-common/src/config.rs</code></li><li>The source side is implemented: xerj autoindex s3://bucket/prefix lists a prefix and streams each changed object into a local mirror — the mirrored object bytes land on local disk under the state directory. — <code>docs/OBJECT_STORAGE.md</code></li>- <li>PUT /_watcher/watch/{id} inserts the body into an in-memory map and answers condition met true; no code evaluates a stored watch. — <code>engine/crates/xerj-api/src/es_compat.rs</code></li>- <li>The console's .xerj_alert_rules and .xerj_alert_fires indices have schemas and are created at bootstrap; no evaluator reads or writes them. — <code>engine/crates/xerj-console-api/src/indices.rs</code></li>+ <li>Since v1.0.0-rc.80, PUT /_watcher/watch/{id} either accepts a watch it will evaluate (interval trigger, one-index input.search, condition.xerj_decide, index_alert action — a background task prefilters with the watch's query, runs /_decide per rendered question, writes .xerj_alert_fires on a non-abstaining win at >= p_min) or refuses the body with a 501 naming what is not evaluated. An earlier version of this claim said no code evaluates a stored watch — true before rc.80, false since. — <code>engine/crates/xerj-api/src/es_compat.rs</code></li>+ <li>.xerj_alert_fires is written by the _watcher evaluator since rc.80 (fire records carry the RAW decide confidence plus p_cal when a calibration is fitted); .xerj_alert_rules still has no writer — the console does not author watches. — <code>engine/crates/xerj-console-api/src/indices.rs</code></li><li>Ingest transforms are built-in native Rust plugins; xerj-wasm has no wasmtime dependency and no wasm feature. — <code>engine/crates/xerj-wasm/Cargo.toml</code></li><li>The xerj-logs crate is compiled in as a dependency and is not wired: it has zero call sites in non-test code. — <code>ROADMAP.md</code></li><li>The roadmap section that carries these statuses, and the design page behind it. — <code>docs/ZERO_TOKEN_DIRECTION.md</code></li>diff --git a/landing/answers/does-xerj-support-s3-alerting-plugins.md b/landing/answers/does-xerj-support-s3-alerting-plugins.mdindex 84cbd9e77..34eae1784 100644--- a/landing/answers/does-xerj-support-s3-alerting-plugins.md+++ b/landing/answers/does-xerj-support-s3-alerting-plugins.md@@ -2,25 +2,25 @@---title: "Does XERJ support S3, alerting or custom plugins?"canonical: "https://xerj.org/answers/does-xerj-support-s3-alerting-plugins"-updated: "2026-09-26"+updated: "2026-09-30"source: "content/answers/does-xerj-support-s3-alerting-plugins.md"---# Does XERJ support S3 storage, alerting or custom plugins yet?-**TL;DR** — One of the three shipped in v1.0.0-rc.77: an index can live in an S3-compatible bucket (`storage.backend = "s3"`). Alerting and user-supplied plugins are still not implemented, and each has a surface in the code that looks closer to done than it is, so this page says exactly what exists, with the file that proves it. The authoritative list is [`ROADMAP.md`](https://github.com/xerj-org/xerj/blob/main/ROADMAP.md).+**TL;DR** — The scoreboard moved twice since this page was written: an index can live in an S3-compatible bucket since v1.0.0-rc.77, record corpus packs shipped in rc.78, and alerting became PARTIAL in rc.80 (`_watcher` evaluates exactly one watch shape and refuses every other loudly). User-supplied ingest plugins remain built-in-only. This page says exactly what exists, with the file that proves it. The authoritative list is [`ROADMAP.md`](https://github.com/xerj-org/xerj/blob/main/ROADMAP.md).## Agent prompt```text-Act as a coding agent. Read https://xerj.org/llms.txt and https://github.com/xerj-org/xerj/blob/main/ROADMAP.md before you plan around XERJ. Object storage works in BOTH directions: xerj autoindex s3://bucket/prefix reads documents out of a bucket, and since v1.0.0-rc.77 storage.backend = "s3" stores the index itself in one (needs an existing storage.s3_bucket; the WAL stays local). Do not design for alerting, user-supplied ingest plugins or downloadable corpus packs: none of those is implemented. Tell the human which of their requirements falls on that list, and use only what the roadmap lists under shipping today.+Act as a coding agent. Read https://xerj.org/llms.txt and https://github.com/xerj-org/xerj/blob/main/ROADMAP.md before you plan around XERJ. Object storage works in BOTH directions: xerj autoindex s3://bucket/prefix reads documents out of a bucket, and since v1.0.0-rc.77 storage.backend = "s3" stores the index itself in one (needs an existing storage.s3_bucket; the WAL stays local). Alerting is PARTIAL (v1.0.0-rc.80: interval-triggered xerj_decide watches that index fires; every other shape 501s). Record packs are real since rc.78 (pre-indexed packs are not). Do not design for user-supplied ingest plugins or pre-indexed packs. Tell the human which of their requirements falls on that list, and use only what the roadmap lists under shipping today.```## Commands### Command 1-Note: The request is accepted and the watch is stored. Nothing will ever evaluate it, and the reply says the condition was met although the watch asked for never.+Note: Since rc.80 this exact body is REFUSED with a 501 naming what is not evaluated (the condition is not xerj_decide). A supported shape — interval trigger, one-index search input, xerj_decide condition, index_alert action — is accepted AND evaluated.```shcurl -s -XPUT http://127.0.0.1:9200/_watcher/watch/demo -H 'content-type: application/json' -d '{"trigger":{"schedule":{"interval":"10s"}},"input":{"simple":{}},"condition":{"never":{}}}'@@ -28,7 +28,7 @@ curl -s -XPUT http://127.0.0.1:9200/_watcher/watch/demo -H 'content-type: applic### Command 2-Note: The stored body comes back unchanged. That is all this API does today.+Note: The stored body comes back unchanged.```shcurl -s -XGET http://127.0.0.1:9200/_watcher/watch/demo@@ -44,9 +44,9 @@ curl -s -XPOST http://127.0.0.1:9200/detections/_search -H 'content-type: applic## Why this page exists-A search engine that speaks a familiar wire protocol invites assumptions. An agent that sees a `_watcher` route may plan around it, and it should not: XERJ does not run watches. An operator who sees `backend = "s3"` in a config schema may expect a bucket to work — since v1.0.0-rc.77 it does, and this page is the one place that used to say otherwise.+A search engine that speaks a familiar wire protocol invites assumptions. An agent that sees a `_watcher` route may plan around full Elasticsearch watcher semantics — cron schedules, transforms, webhooks — and it should not: since v1.0.0-rc.80 XERJ evaluates exactly one watch shape and refuses the rest. An operator who sees `backend = "s3"` in a config schema may expect a bucket to work — since v1.0.0-rc.77 it does, and this page is the one place that used to say otherwise.-XERJ's rule is that an input is either honoured or refused loudly. Two of the three follow that rule today and one does not (`_watcher` accepts a watch and never evaluates it — the accepted-and-ignored surface). All of them were checked against `main` by reading the named file: 2026-09-18 originally, with the S3 status re-verified on 2026-09-26 against v1.0.0-rc.77.+XERJ's rule is that an input is either honoured or refused loudly. `_watcher` was the one surface that broke the rule (accepted a watch, never evaluated it); v1.0.0-rc.80 closed that — [#1082](https://github.com/xerj-org/xerj/pull/1082), closing [#1062](https://github.com/xerj-org/xerj/issues/1062). All statuses on this page were checked against `main` by reading the named file: 2026-09-18 originally, S3 re-verified 2026-09-26 against v1.0.0-rc.77, alerting and packs re-verified 2026-09-30 against v1.0.0-rc.80.## S3 and object storage: both directions work@@ -64,15 +64,17 @@ This section said the opposite until 2026-09-26 — it described `S3Backend` asOne refusal remains, and it is the loud kind: with `storage.backend = "s3"` and no `storage.s3_bucket`, the server does not start. It prints that `storage.backend = "s3"` requires `storage.s3_bucket` to name an existing bucket — XERJ never creates one. That check is in `engine/crates/xerj-common/src/config.rs`.-## Alerting: there is none+## Alerting: one real watch shape since v1.0.0-rc.80-XERJ has no alerting. Two surfaces make it look otherwise.+This section said "there is none" until 2026-09-30 — and that was true when it was checked (2026-09-18): `PUT /_watcher/watch/{id}` accepted every body, replied that the condition was met, and no code ever evaluated anything. v1.0.0-rc.80 ([#1082](https://github.com/xerj-org/xerj/pull/1082)) replaced the accepted-and-ignored surface with evaluate-or-refuse.-The first is `_watcher`. `PUT /_watcher/watch/{id}` is accepted. The handler puts the body into an in-memory map and replies that the condition was met, whatever condition you sent. `GET` and `DELETE` work on the same map. There is no scheduler, and no code ever evaluates a stored watch. This one is an accepted-and-ignored input, and the roadmap says it must either run watches or refuse them.+**What evaluates.** Exactly one watch shape: `trigger.schedule.interval` (>= 1s), `input.search` over exactly one index, `condition.xerj_decide` (a question, a positive label, `p_min`, `k`, an optional decide index) and an `index_alert` action. A per-watch background task prefilters candidates with the watch's own query (ascending `_seq_no` keyset paging, 100 per page), runs each rendered question through the real `/_decide`, and writes a fire record into `.xerj_alert_fires` when the positive label wins non-abstaining at or above `p_min`. Fire records carry the RAW decide confidence, plus `p_cal` with `calibrated: true` when the node has a fitted calibration — the threshold stays on the raw value.-The second is the console. It owns two system indices, `.xerj_alert_rules` and `.xerj_alert_fires`. Both have schemas and both are created at start-up. No evaluator reads the first or writes the second, so there are no alerts and no notifications from them either.+**What refuses.** Everything else — cron schedules, transforms, other action types, multi-index input — is a 501 that names what is not evaluated. Malformed supported shapes are 400.-One piece underneath is real. The `percolate` query is a dispatched query type: you store queries as documents, and a `percolate` search returns the stored queries that match a document you supply. A future detection feature would use that as its cheap first stage. The stages after it, a typed judgment and an alert that carries a calibrated probability, are planned and do not exist.+**Honest limits.** A restart does not resume watches (evaluators are spawned by `put_watch` only); the per-pass decide budget is 1000, with the cursor held below a failure so the backlog is retried, never dropped; the console's `.xerj_alert_rules` index still has no writer — watches are authored through the API, not the UI.++The `percolate` query underneath is unchanged and real: you store queries as documents, and a `percolate` search returns the stored queries that match a document you supply. The watcher's prefilter is that idea, productised for one condition type.## Custom ingest plugins: built-in only@@ -86,7 +88,7 @@ The refusal rule holds here. A pipeline that names a processor this build does n**A log-specific index mode.** The `xerj-logs` crate is in the workspace and is compiled in as a dependency, but it is not wired: it has zero call sites outside its own tests. Logs you index today go through the general segment format and the ordinary aggregations. The [low-volume log search page](/answers/cheap-low-volume-log-search) describes what does work.-**Downloadable corpus packs.** A hub of signed, pre-indexed packs is on the roadmap and no code exists. The design work is the pack format: a format version that readers refuse when they do not know it, per-file checksums, a signature, and licence and provenance on every record. Three risks decide whether it ships: whether the source licence allows redistribution, whether a pack is safe to open, and whether the format can stay stable after packs are published.+**Downloadable corpus packs — the RECORD half shipped in v1.0.0-rc.78** ([#1046](https://github.com/xerj-org/xerj/pull/1046), [#1048](https://github.com/xerj-org/xerj/pull/1048); this section said "no code exists" before, which was true then). `xerj corpus build` turns a declarative recipe into a checksummed portable pack of records — a `format_version` readers refuse when unknown, licence and provenance on every record — `corpus add --from <pack>` verifies the per-file SHA-256 and `--verify-sig` a detached ed25519 signature, and the `rust-vulns` pack is rebuilt, signed and published daily as a dated GitHub Release. Still unbuilt: the PRE-INDEXED half (an already-indexed bundle you mount instead of indexing records yourself) and a hub directory beyond that one pack.## What to do instead today@@ -94,13 +96,13 @@ The refusal rule holds here. A pipeline that names a processor this build does n| --- | --- || index data in S3 | works since v1.0.0-rc.77: `storage.backend = "s3"` plus an existing `storage.s3_bucket` — segment bundles and `snapshot.json` live in the bucket, the WAL stays local || search documents that live in S3 | works: `xerj autoindex s3://bucket/prefix` mirrors the changed objects to local disk and indexes them |-| an alert when a document matches | there is no alerting; run a `percolate` or an ordinary search on your own schedule and act on the result yourself |+| an alert when a document matches | one shape works since rc.80: an interval watch with a `xerj_decide` condition over one index writes fires into `.xerj_alert_fires`; there is no notification delivery and no console authoring yet || a custom transform at ingest | one of the built-in transforms, or transform the document before you send it |-| a ready-made reference corpus | clone the source and run `xerj autoindex` on it |+| a ready-made reference corpus | `xerj corpus add --from` a record pack (the daily `rust-vulns` pack is published), or clone source and run `xerj autoindex` on it |## What this page does not claim-It does not claim a date for the two that remain — alerting and user-supplied plugins are planned, which means no code exists yet. S3 as an index home shipped in v1.0.0-rc.77; before that release the same setting refused to start, and this page said so.+It does not claim more alerting than one watch shape: no cron, no transforms, no notification delivery, no watch authoring in the console, and watches do not survive a restart. Those remain planned. It does not claim pre-indexed packs or a pack hub: only record packs exist. S3 as an index home shipped in v1.0.0-rc.77; before that release the same setting refused to start, and this page said so.It does not claim the search side is limited in the same way. Full-text search, the `hybrid` query and vector search are shipping, and the default embedder is lexical feature hashing, not a neural model. The [hybrid retrieval page](/answers/how-xerj-combines-search) covers that.@@ -110,7 +112,7 @@ XERJ is single-node. Everything on this page describes one process on one host.### Can XERJ store its index in S3, send alerts, or run my own ingest plugin?-One of the three, since v1.0.0-rc.77: an index can live in an S3-compatible bucket (`storage.backend = "s3"`). Alerting and user-supplied plugins are still not implemented, and each has a surface in the code that looks closer to done than it is. This page says exactly what exists.+S3 fully, since v1.0.0-rc.77 (`storage.backend = "s3"`); alerting PARTIALLY, since v1.0.0-rc.80 (one watch shape evaluates, everything else 501s); record corpus packs since rc.78. User-supplied ingest plugins are the one still not implemented, and the `xerj-wasm` crate name makes it look closer to done than it is. This page says exactly what exists.### Does XERJ support S3 or object storage?@@ -118,7 +120,7 @@ Yes, in both directions. `xerj autoindex s3://bucket/prefix` reads documents OUT### Does XERJ have alerting or a working watcher?-No. XERJ has no alerting. `_watcher` stores watches and no scheduler ever evaluates them, and the console alert-rule indices have schemas and no evaluator.+Partially, since v1.0.0-rc.80. One watch shape is real: `PUT /_watcher/watch/{id}` with an interval trigger, a one-index search input, a `condition.xerj_decide` question and an `index_alert` action runs a background evaluator that writes fires into `.xerj_alert_fires` when the positive label wins non-abstaining at or above `p_min`. Everything else — cron, transforms, other actions, multi-index input — is refused with a 501 that names what is not evaluated. Restarts do not resume watches.### Can I write my own ingest plugin for XERJ?@@ -134,7 +136,7 @@ Yes, one piece. The `percolate` query is a dispatched query type and matches sto### Can I download a pre-indexed corpus for XERJ?-No. A hub of signed, pre-indexed packs is planned and no code exists. The open questions are redistribution licence, pack safety and format stability, and they decide whether it ships.+Record packs, yes, since v1.0.0-rc.78: `xerj corpus build` turns a recipe into a checksummed, signable pack of records and `corpus add --from <pack>` verifies it — the rust-vulns pack is published daily. PRE-INDEXED packs (an indexed bundle you mount) are still unbuilt, and there is no hub beyond that one pack.### Where is the authoritative status?@@ -145,8 +147,8 @@ No. A hub of signed, pre-indexed packs is planned and no code exists. The open q- storage.backend = "s3" stores the index in the bucket: one immutable ZBM1 bundle object per segment family plus a per-index snapshot.json catalogue; merges publish before retiring inputs; a fresh node adopts the bucket. — `docs/OBJECT_STORAGE.md`- The one remaining startup refusal is the loud kind: storage.backend = "s3" requires storage.s3_bucket to name an existing bucket — XERJ never creates one. — `engine/crates/xerj-common/src/config.rs`- The source side is implemented: xerj autoindex s3://bucket/prefix lists a prefix and streams each changed object into a local mirror — the mirrored object bytes land on local disk under the state directory. — `docs/OBJECT_STORAGE.md`-- PUT /_watcher/watch/{id} inserts the body into an in-memory map and answers condition met true; no code evaluates a stored watch. — `engine/crates/xerj-api/src/es_compat.rs`-- The console's .xerj_alert_rules and .xerj_alert_fires indices have schemas and are created at bootstrap; no evaluator reads or writes them. — `engine/crates/xerj-console-api/src/indices.rs`+- Since v1.0.0-rc.80, PUT /_watcher/watch/{id} either accepts a watch it will evaluate (interval trigger, one-index input.search, condition.xerj_decide, index_alert action — a background task prefilters with the watch's query, runs /_decide per rendered question, writes .xerj_alert_fires on a non-abstaining win at >= p_min) or refuses the body with a 501 naming what is not evaluated. An earlier version of this claim said no code evaluates a stored watch — true before rc.80, false since. — `engine/crates/xerj-api/src/es_compat.rs`+- .xerj_alert_fires is written by the _watcher evaluator since rc.80 (fire records carry the RAW decide confidence plus p_cal when a calibration is fitted); .xerj_alert_rules still has no writer — the console does not author watches. — `engine/crates/xerj-console-api/src/indices.rs`- Ingest transforms are built-in native Rust plugins; xerj-wasm has no wasmtime dependency and no wasm feature. — `engine/crates/xerj-wasm/Cargo.toml`- The xerj-logs crate is compiled in as a dependency and is not wired: it has zero call sites in non-test code. — `ROADMAP.md`- The roadmap section that carries these statuses, and the design page behind it. — `docs/ZERO_TOKEN_DIRECTION.md`diff --git a/landing/answers/index.html b/landing/answers/index.htmlindex 0c510e870..4b9e69740 100644--- a/landing/answers/index.html+++ b/landing/answers/index.html@@ -63,7 +63,7 @@"url": "https://xerj.org/answers/","image": "https://xerj.org/og/xerj-card.png","inLanguage": "en",- "dateModified": "2026-09-26",+ "dateModified": "2026-09-30","isPartOf": {"@id": "https://xerj.org/#website"},@@ -261,7 +261,7 @@</ul><h2>Capabilities and limits</h2><ul>- <li><a href="/answers/does-xerj-support-s3-alerting-plugins">Does XERJ support S3 storage, alerting or custom plugins yet?</a> — Since v1.0.0-rc.77 the index CAN live in an S3-compatible bucket (storage.backend = "s3": one immutable bundle object per segment family plus snapshot.json). _watcher never runs a watch; plugins are built-in.</li>+ <li><a href="/answers/does-xerj-support-s3-alerting-plugins">Does XERJ support S3 storage, alerting or custom plugins yet?</a> — Since v1.0.0-rc.77 the index CAN live in an S3-compatible bucket; since rc.78 record packs are real; since rc.80 _watcher evaluates exactly one watch shape and refuses every other with 501. Plugins stay built-in.</li></ul><h2>Code security</h2><ul>diff --git a/landing/answers/index.json b/landing/answers/index.jsonindex 6d6b20885..7d1a1770c 100644--- a/landing/answers/index.json+++ b/landing/answers/index.json@@ -98,8 +98,8 @@"canonical_url": "https://xerj.org/answers/does-xerj-support-s3-alerting-plugins","markdown_url": "https://xerj.org/answers/does-xerj-support-s3-alerting-plugins.md","cluster": "Capabilities and limits",- "updated": "2026-09-26",- "summary": "Since v1.0.0-rc.77 the index CAN live in an S3-compatible bucket (storage.backend = \"s3\": one immutable bundle object per segment family plus snapshot.json). _watcher never runs a watch; plugins are built-in."+ "updated": "2026-09-30",+ "summary": "Since v1.0.0-rc.77 the index CAN live in an S3-compatible bucket; since rc.78 record packs are real; since rc.80 _watcher evaluates exactly one watch shape and refuses every other with 501. Plugins stay built-in."},{"title": "Audit a PHP codebase with an index, not grep",diff --git a/landing/sitemap.xml b/landing/sitemap.xmlindex f936805bf..20c14ee2b 100644--- a/landing/sitemap.xml+++ b/landing/sitemap.xml@@ -2,11 +2,11 @@<urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9"><url><loc>https://xerj.org/</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/agent-search/</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/aise-demo</loc>@@ -338,19 +338,19 @@</url><url><loc>https://xerj.org/blog/cutting-the-index-28-percent</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/blog/does-xerj-beat-jev</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/blog/jev-vs-a-bm25-vote</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/blog/one-vulnerability-three-names</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/blog/the-rc80-gates</loc>@@ -458,115 +458,115 @@</url><url><loc>https://xerj.org/docs/</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/agents/</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/agents/endpoints</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/agents/quickstart</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/aggregations</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/analyzers</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/api-es-compat</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/api-native</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/backup-restore</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/cli</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/clustering</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/compression</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/config</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/env</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/ingest</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/install</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/metrics</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/migration-from-es</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/operations</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/playbooks/full-text</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/playbooks/log-analytics</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/playbooks/observability</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/playbooks/siem</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/playbooks/vector-search</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/queries</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/quickstart</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/recipes/</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/recipes/agentic-memory</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/recipes/air-gapped-deployment</loc>@@ -574,71 +574,71 @@</url><url><loc>https://xerj.org/docs/recipes/anomaly-detection</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/recipes/continuous-anomaly-datafeeds</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/recipes/document-folder-index</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/recipes/hybrid-search</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/recipes/log-analytics</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/recipes/migrate-from-elasticsearch</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/recipes/passage-retrieval</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/recipes/semantic-search-rag</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/recipes/vector-quantization</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/recipes/vector-search-knn</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/recipes/zero-config-autoindex</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/security</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/storage</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/troubleshooting</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/upgrades</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/docs/vectors</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/for-agents</loc>- <lastmod>2026-09-29</lastmod>+ <lastmod>2026-09-30</lastmod></url><url><loc>https://xerj.org/industries/</loc>diff --git a/scripts/seo/claims_rules.py b/scripts/seo/claims_rules.pyindex 767530699..9e43022a5 100644--- a/scripts/seo/claims_rules.py+++ b/scripts/seo/claims_rules.py@@ -776,13 +776,32 @@ RULES = ["pattern": r"(?:\balert(?:s|ing)\b|\b_watcher\b|\bwatches\b|\bnotif(?:y|ication)\w*\b\s*(?:rule|when|on)|"r"\bpage(?:s|rduty)?\b[^.\n]{0,20}when|trigger\w*[^.\n]{0,25}(?:alert|notification))","exempt": [_neg_near(r"alert\w*|_watcher|watches|notification\w*", 140),- r"no scheduler|0/7|stores watches"],- "reason": "`_watcher` stores watches but NO SCHEDULER executes them. There is no rule "- "engine and no notification path. Kibana alerting scores 0/7 DELIVERED.",- "evidence": [LC + ":264", LC + ":635", RJ + ":27"],- "rewrite": "'`_watcher` accepts and stores watch definitions for client compatibility, but "- "nothing executes them - there is no scheduler and no notification path. Run "- "alerting outside XERJ.'",+ r"no scheduler|0/7|stores watches",+ # Since v1.0.0-rc.80 (#1082) exactly ONE watch shape evaluates+ # (interval trigger, one-index search input, condition.xerj_decide,+ # index_alert action -> .xerj_alert_fires) and every other shape is+ # refused with a 501 naming what is not evaluated. Mentions qualified+ # by that partial status are the honest framings; an UNQUALIFIED+ # "alerting rules fire exactly as on Elasticsearch" still fails.+ r"one [a-z ]{0,14}watch shape",+ r"not evaluated|never evaluated|refus\w+[^.\n]{0,60}501|501[^.\n]{0,60}refus",+ r"\bxerj_decide\b",+ r"re-verified[^.\n]{0,80}v1\.0\.0-rc\.80",+ r"PARTIAL"],+ "reason": "Since v1.0.0-rc.80 (#1082) `_watcher` evaluates exactly ONE watch shape "+ "(interval trigger, one-index `input.search`, `condition.xerj_decide`, "+ "`index_alert`) and writes fires to `.xerj_alert_fires`; every other shape "+ "is refused with a 501 naming what is not evaluated. There is still no "+ "notification delivery, no cron, no transforms, no console authoring, and "+ "a restart does not resume watches.",+ "evidence": [LC + ":264", LC + ":635", RJ + ":27",+ "ROADMAP.md:117",+ "CHANGELOG.md:440"],+ "rewrite": "'Alerting is PARTIAL since v1.0.0-rc.80: one watch shape evaluates (interval "+ "trigger + one-index search input + `condition.xerj_decide` + `index_alert`) "+ "and writes fire records; every other shape is a 501 naming what is not "+ "evaluated, and there is no notification delivery. Run delivery outside "+ "XERJ.'",},{"id": "FC-ML",